AGM-style belief management — expand, revise, or contract beliefs with automatic contradiction detection and confidence propagation.
Operation and target: $ARGUMENTS
act(action: "belief_revise", operation: "expand", content: "New fact here", confidence: 0.8)
Automatically checks for contradictions. Creates bidirectional :contradicts edges if found (forming kappa=1 SCCs).
act(action: "belief_revise", operation: "revise", node_id: "<old_node>", content: "Corrected fact", rationale: "why")
Creates new node, supersedes old (confidence -> 30%), propagates 0.6x decay through dependents.
act(action: "belief_revise", operation: "contract", node_id: "<wrong_node>", rationale: "why")
Sets confidence to 0.05, propagates decay. Node preserved for provenance.
learn(action: "contradictions", content: "<new info>")
learn(action: "contradictions", node_id: "<existing_node>")
Returns detected conflicts with similarity scores and negation markers.
learn(action: "contradictions", content: "<new info>") — check first
Assess: is the contradiction real?
act(action: "belief_revise", operation: "revise|contract|expand") — act
route(action: "topology", ...) — check for new cycles
route(action: "deliberate", ...) — if kappa > 0 on affected region
learn(action: "from_outcome", ...) — report the revision result
Always check contradictions before revising
Use revise when old is wrong (not expand — that creates contradictions)
Use contract when wrong but no replacement available
Review affected_node_ids after revision to assess cascade impact
Expanding without checking for contradictions first
Using contract when you have a correction (use revise instead)
Ignoring cascade effects on dependent nodes